You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
🖥️ Computer-Use Agents: An Overview of Terminal-Based and GUI-Based Approaches
A curated list of papers, tools, and benchmarks on computer-use agents: systems that let LLMs operate computers through terminals or graphical interfaces.
We organize this collection into two families:
Terminal / CLI-Based Agents interact with computers through text interfaces: shell commands, file I/O, and code execution. This category spans five stages of development: code completion, chat assistants, autonomous coding agents, general-purpose sandbox agents, and personal AI assistants.
GUI-Based Agents perceive and act on graphical interfaces through mouse clicks, keyboard input, and screen observation. We cover web agents, desktop/OS agents, and mobile agents, each with distinct perception strategies and challenges.
Note: This is a preliminary collection and is by no means exhaustive. Many important papers and tools are likely missing. If you know of work that should be included, please open a PR or issue. Contributions of all kinds are welcome!
We also wrote a short report discussing the development and architecture of these systems: report (PDF).